Publication number | US7565389 B2 |

Publication type | Grant |

Application number | US 10/532,903 |

PCT number | PCT/IB2003/004685 |

Publication date | Jul 21, 2009 |

Filing date | Oct 20, 2003 |

Priority date | Oct 30, 2002 |

Fee status | Paid |

Also published as | CN1708902A, CN100525102C, DE10250555A1, EP1559193A1, US20050289201, WO2004040757A1 |

Publication number | 10532903, 532903, PCT/2003/4685, PCT/IB/2003/004685, PCT/IB/2003/04685, PCT/IB/3/004685, PCT/IB/3/04685, PCT/IB2003/004685, PCT/IB2003/04685, PCT/IB2003004685, PCT/IB200304685, PCT/IB3/004685, PCT/IB3/04685, PCT/IB3004685, PCT/IB304685, US 7565389 B2, US 7565389B2, US-B2-7565389, US7565389 B2, US7565389B2 |

Inventors | Gerhard Runze |

Original Assignee | Nxp B.V. |

Export Citation | BiBTeX, EndNote, RefMan |

Patent Citations (10), Non-Patent Citations (1), Referenced by (4), Classifications (5), Legal Events (5) | |

External Links: USPTO, USPTO Assignment, Espacenet | |

US 7565389 B2

Abstract

In a method for determining filter coefficients of a digital filter, particularly in UMTS, the predetermined filter coefficients bv are divided by a same scaling factor s and then quantized in that counted from the most significant bit onwards only a certain number n of “1” bits is used and in that the quantization error E(s) is minimized by the selection of the scaling factor s0 and these scaled and quantized filter coefficients βv are implemented in the filter.

Claims(19)

1. A method, comprising:

determining a plurality of filter coefficients for a digital filter for a Universal Mobile Telecommunication System (UMTS), the determining including:

dividing initial filter coefficient by a scaling factor, to result in a plurality of scaled filter coefficients:

quantizing the scaled filter coefficient so that only a certain maximum number (n) of “1” bits are counted from a most significant bit onwards:

determining a respective quantization error of each quantized scaled filter coefficient relative to a respective one of the initial filter coefficients; and

repeatedly modifying for a number of times the scaling factor and determining which scaling factor results in a quantization error having minimal error value, and

implementing in the filter the filter coefficients resulting in the minimal error value.

2. A method as claimed in claim 1 , wherein the number (n) comprises one of four, three, or two.

3. A method as claimed in claim 1 , wherein if again a “1” bit follows the last “1” bit, a rounding is effected from the last bit onwards.

4. The new method according to claim 1 , further comprising multiplying an input by 2^{j }by shifting the input by i and summing shifted values using a plurality of adders.

5. The method according to claim 4 , wherein the multiplying is performed using a plurality of multipliers, the method further comprising selectively connecting the multipliers with the adders using a programmable selector in accordance with a programming.

6. The method according to claim 1 , further comprising:

multiplying an input by 2^{j }by shifting the input by i using a multiplier formed by connections of inputs and outputs.

7. A digital filter for a Universal Mobile Telecommunication System (UMTS), comprising

means for dividing a plurality of binary filter coefficients by a scaling factor to result in a plurality of scaled filter coefficients;

means for quantizing the scaled filter coefficients so that they do not exceed a selected number (n) of “1” bits from a most significant bit onwards; and

adder stages for processing the scaled and quantized filter coefficients with an input signal.

8. The digital filter as claimed in claim 7 , comprising a final stage for processing an output signal by a factor reciprocal to the scaling factor.

9. A digital filter as claimed in claim 7 , wherein each adder stage comprises n−1 adders and means for multiplying an input by 2^{i }by shifting the input by i, the input being a respective one of the scaled and quantized filter coefficients.

10. A digital filter as claimed in claim 7 , wherein the adder stages include first and second adder stages respectively including:

respective numbers of multiplying means for multiplying an input by 2^{i }by shifting the input by i, wherein the respective numbers of multiplying means are different: and

respective numbers of adders coupled to the respective multiplying means, wherein the respective numbers of adders are different.

11. A digital filter as claimed **10**, wherein the adder stages include a third adder stage having only a single multiplying means for multiplying an input by 2^{j }by shifting the input by i.

12. A digital filter as claimed in claim 7 , wherein each adder stage comprises n-1 adders and means for multiplying an input by 2^{j }by shifting the input by i, the input being a respective one of the scaled and quantized filter coefficients, wherein each of the means for multiplying the input by 2^{i }by shifting the input by i is formed by connections of inputs and outputs of a multiplier stage.

13. A digital filter as claimed in claim 7 , wherein each adder stage comprises:

n-1 adders;

multiplying means for multiplying an input by 2^{j }by shifting the input by j, the input being a respective one of the scaled and quantized filter coefficients; and

a programmable selector which in accordance with programming connects the multiplying means with the adders.

14. A method, comprising:

producing a plurality of scaled filter coefficients by dividing initial filter coefficients by a scaling factor;

quantizing the scaled filter coefficients so that only a certain maximum number (n) of “1” bits are counted from a most significant bit onwards;

determining respective quantization errors of the quantized scaled filter coefficients relative to the initial filter coefficients, respectively; and

modifying the scaling factor and determining which scaling factor results in a quantization error having minimal error value, and

implementing in the filter the filter coefficients resulting in the minimal error value.

15. A method as claimed in claim 14 , wherein the number (n) comprises one of four, three, or two.

16. A method as claimed in claim 14 , wherein if again a “1” bit follows the last “1” bit, a rounding is effected from the last bit onwards.

17. The method according to claim 14 , further comprising:

multiplying an input by 2^{j }by shifting the input by i using a multiplier formed by connections of inputs and outputs.

18. The method according to claim 14 , further comprising

multiplying an input by 2^{j }by shifting the input by i and summing shifted values using a plurality of adders.

19. The method according to claim 18 , wherein the multiplying is performed using a plurality of multipliers, the method further comprising selectively connecting the multipliers with the adders using a programmable selector in accordance with a programming.

Description

The invention relates to a method for modifying filter coefficients for a digital filter, more particularly for UMTS (Universal Mobile Telecommunication System), in which the filter coefficients are predetermined and modified in a filter design program. Furthermore, the invention relates to a digital filter of this type.

Digital filters are used, for example, as pulse shaping filters and matched filters in the UMTS standard (3GPP, TS 25.201: UE Radio Transmission and Reception (FDD), V3.2.0) and (3GPP, TS 25.213: Spreading and Modulation (FDD), V3.0.0). Owing to the number of filter coefficients to be processed and the large signal bandwidth such filters require much computation circuitry. Therefore, they can hardly be implemented by a signal processor which is part of the appliance that has the filter function. A hardware implementation necessary as a result and having its own components (chips) for the filters leads to much surface area in the appliance and to increased energy consumption. These two elements are thwarting a compact construction of the appliance.

In U.S. Pat. No. 6,311,203 is described a digital filter in which two factors are multiplied by to simplify the calculation of the filter coefficients. A preselection of filter coefficients that can be evaluated in a simple manner is not provided.

In U.S. Pat. No. 5,732,004 scaling factors are assigned to individual filter coefficients. A common scaling factor for all the filter coefficients is not provided.

From WO 01/22 582 A1 is known a floating point FIR filter. A common scaling factor is not provided.

It is an object of the invention to provide a method and a filter of the type defined in the opening paragraph in which without a decisive deterioration of the filter properties and of the processing rate, the filter is produced with a small chip surface and little current consumption.

The above object with respect to the method is achieved by the characteristic features of the present invention and with respect to the characteristic features of a filter according to the present invention.

For this purpose, the ideal filter coefficients previously determined in a filter design program are quantized and scaled so that they can be processed by a simple adding operation and costly multipliers are avoided. The filter coefficients simplified by quantization and scaling are not determined in the respective appliance that comprises the filter function. They are determined from the ideal filter coefficients in an extended filter design program and subsequently implemented in the appliance.

The quantized and scaled filter coefficients simplify the numerical complexity of the computation of the filtered signal. This allows to realize the filter function that has a small required area or space in the appliance and with little energy consumption. The filter may be an up-link and/or down-link, root-raised cosine filter of the UMTS with fixed-point arithmetic. The described measures, however, may also be used in other non-recursive filters working with fixed-point arithmetic.

Advantageous embodiments of the invention are apparent from the following description of the present invention. In the drawing:

In prior-art filters (_{N−1 }to b_{0 }in a plurality of multipliers **1**. The filter coefficients are predetermined by means of a filter design program corresponding to the desired type of filter. The resulting partial signals Y_{N−1 }to Y_{0 }are summed by adders **2** in a time-delayed manner (z^{−1}), the sum denoting the filtered output signal V. The ideal frequency response is shown in a dashed line in the

In the described filter, coefficients which are simplified by scaling and quantization are used which can be processed in a simpler manner. The flow chart of _{v }are formed from the predetermined filter coefficients b_{v}. This is effected prior to the implementation of the filter coefficients in the filter.

Once the filter coefficients b_{v }typically of a filter have first been determined by a standard filter design, they are divided by a scaling factor which is the same for all the filter coefficients. The result is the scaled filter coefficients β_{v}=b_{v}/s.

Subsequently, a quantization of the scaled filter coefficients takes place. During the quantization the number n of the “1” bits found after the most significant bit (MSB) is limited to a certain maximum number n, where n is 4, 3 or 2 in the examples described. The number n may be different for each adder stage depending on the coefficients provided. This is based on the recognition that a multiplication of a signal value by a filter coefficient having few “1” bits needs fewer adding operations than a multiplication by an arbitrary filter coefficient of a same effective word length.

As a result of the quantization, distortions of the filtered signal will inevitably occur. This quantization error is smallest possible.

After the quantization the respective quantization error:

*E*(*s*)=Σ|*b* _{v} *−sx β* _{v}|^{2 }

is determined (compare _{step}) and the quantization error occurring with this new scaling factor is again determined and stored. If then the scaling factor increased in steps exceeds 2.0, a comparison is made what scaling factor s_{0 }leads to a minimum quantization error. This scaling factor s_{0 }is then used for the final determination of the filter coefficient β_{v }found via scaling and quantization, where

β_{v} *=Q*(*b* _{v} */s* _{0})

Q then describes the quantization.

At the output of the filter the division of the original filter coefficient by the common value s_{0 }is canceled as a result of a multiplication by the common scaling factor s_{0}.

The following Table shows stages of the quantization algorithm in the light of a numerical example:

Stage | Binary | Decimal | quantization error |

original value | 0.101101111 | 0.716796875 | — |

step 1:4 “1” bits | 0.101101 | 0.703125 | 0.013671875 |

consider next bit | 0.1011011 | 0.7109375 | 0.005859375 |

round at fourth “1” bit | 0.101110 | 0.71875 | 0.00195125 |

In the Table it is assumed that the number n of “1” bits allowed after the most significant bit is n=4.

As appears from the Table the binary original value has seven “1” bits. In accordance with the guideline, in step 1 the binary value is limited to four “1” bits after the most significant bit, here 0. This corresponds to a decimal value of 0.703125, which as against the decimal original value means a quantization error of 0.013671875.

In a second step the next bit following the last “1” bit of step 1 is considered, which again is a “1” bit in the example. Since this bit is a “1” bit, a rounding is made in the next step, so that the binary value 0.1011110 arises. This corresponds to a decimal value 0.71875 which, compared to the decimal original value—irrespective of the scaling factor used—means a quantization error of −0.00195125. The rounding thus considerably reduces the quantization error.

_{0}, k_{0}, l_{0 }and m_{0 }having the values 1, 3, 4 and 5. This processing requires only four steps i_{0}, k_{0}, l_{0}, m_{0}. Each multiplication by this “1” corresponds to a shift of the input signal, here decimal: 217, binary 11011001. Accordingly, i_{0 }is a shift by one position, k_{0 }is a shift by three positions, l_{0 }a shift by four positions, m_{0 }a shift by five positions. Such shifts can be realized with little circuitry because only data lines need to be properly connected for this purpose, as is shown by

As a result of the described quantization, the optimum scaling factor s_{0 }can be sought in a limited interval, for example, in the interval from 1 to 2 or in the interval from ½√2 to √2.

Said filter coefficients β_{v }allow a simplified filter structure or a simplified implementation of the filter, respectively, because multipliers are replaced by adders.

**1** of **3** in which the scaled and quantized filter coefficients β_{v}, β_{N−1 }to β_{0 }respectively, are processed. In a final stage **4**, which may be arranged as a multiplier, a sum signal is processed with a factor s_{0 }reciprocal to the scaling factor of the filter coefficient 1/s_{0}, to do away with the effect the scaling factor has on the output signal V. The final stage **4** may be omitted if the absolute value does not play a decisive role.

**3** for the filter coefficient b_{N−1}. **5**, **6**, **7**, **8** are provided which are designed for the multiplication by squares, that is, −i_{N−1}, −k_{N−1}, −l_{N−1}, −m_{N−1 }in accordance with the respective filter coefficient, while these powers correspond to the example i_{0}, k_{0}, l_{0}, m_{0 }of

The result of the multipliers **5** and **6** is summed in an adder **9**. The result of the multipliers **7** and **8** is summed in an adder **11**. The results of the two adders are added in a further adder **10** to an intermediate signal Y_{N−1}. The number of adders **9**, **10**, **11** is n−1 in this case and in the other examples of embodiment. If n=1, only one multiplier and no adder is necessary.

In the example of embodiment shown in **5**, **6** and one adder **10** are necessary.

In the example of embodiment shown in **3** comprises a programmable selector **12**. Due to the programmable structure the filter coefficients need not of necessity be determined prior to the implementation of the filter. Rather a general programmable filter can be implemented with the selectors, the filter coefficients of which have a maximum of n “1” bits. It is again assumed here that n=4 (compare **9**, **10**, **11** are necessary. By means of the programmable selector **12** its outputs X_{i}, X_{k}, X_{l}, X_{m }are selected accordingly. The range of values of the inputs i, k, l and m is situated between 0 and M−1 and connects one of the respective inputs X_{0 }. . . X_{M−1 }with one of the outputs. The output X_{i }is determined by i, x_{m }is determined by m and so on. For example, a programming of the inputs with i=1, k=3, l=4 and m=5 selects the outputs at X_{i}=X_{l}, X_{k}=X_{3}, X_{l}=X_{4 }and X_{m}=X_{5}. M in this example is 8 (compare

**5**, **6**, **7**, **8**. This multiplier is formed by a simple connection of the data lines between the 8-bit input and the 8-bit output.

_{0}.

In all examples of embodiment the quantization and optimization of the scaling factor has achieved that instead of costly multipliers (**3** can be used which reduces the required chip area of the hardware. The multipliers **5** to **8** of the adder stage **3** can be manufactured in a considerably simpler fashion, contrary to the multipliers of **3** are carried out simultaneously and by equally many adders.

When the proposed quantization is used in a UMTS Root-Raised Cosine Filter, it is possible in a filter with symmetrical coefficients of length **32** to replace the 16 12-bit multipliers **1** with 16 12-bit adder stages **3**.

Patent Citations

Cited Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|

US4791597 | Oct 27, 1986 | Dec 13, 1988 | North American Philips Corporation | Multiplierless FIR digital filter with two to the Nth power coefficients |

US5548540 * | Jun 24, 1994 | Aug 20, 1996 | General Electric Company | Decimation filter having a selectable decimation ratio |

US5732004 | Nov 14, 1995 | Mar 24, 1998 | Advanced Micro Devices, Inc. | DSP architecture for a FIR-type filter and method |

US5831880 * | Sep 20, 1996 | Nov 3, 1998 | Samsung Electronics Co., Ltd. | Method for processing a signal in a CSD filter and a circuit therefor |

US6243729 * | Dec 31, 1998 | Jun 5, 2001 | Texas Instruments Incorporated | Digital finite-impulse-response (FIR) filter with a modified architecture based on high order Radix-N numbering |

US6311203 | Dec 2, 1998 | Oct 30, 2001 | Mitsubishi Denki Kabushiki Kaisha | Multiplier, and fixed coefficient FIR digital filter having plural multipliers |

US6505221 * | Sep 20, 1999 | Jan 7, 2003 | Koninklijke Philips Electronics N.V. | FIR filter utilizing programmable shifter |

EP0766388A2 | Sep 13, 1996 | Apr 2, 1997 | Samsung Electronics Co., Ltd. | Method for processing signal in CSD filter and circuit suitable for the method |

EP0766388B1 | Sep 13, 1996 | Dec 7, 2005 | Samsung Electronics Co., Ltd. | Method for processing signal in CSD filter and circuit suitable for the method |

WO2001022582A1 | Aug 18, 2000 | Mar 29, 2001 | Koninklijke Philips Electronics Nv | Fir filter utilizing programmable shifter |

Non-Patent Citations

Reference | ||
---|---|---|

1 | Henry Samueli; An Improved Search Algorithm for the Design of Multiplierless Coefficients; vol. 36; No. 7; Jul. 1, 1989; pp. 1044-1047; New York, US. |

Referenced by

Citing Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|

US7809393 * | May 6, 2004 | Oct 5, 2010 | Nxp B.V. | Method and arrangement for setting the transmission of a mobile communication device |

US7921147 * | Jul 11, 2007 | Apr 5, 2011 | Sanyo Electric Co., Ltd. | Filtering integrated circuit |

US20080016136 * | Jul 11, 2007 | Jan 17, 2008 | Sanyo Electric Co., Ltd. | Filtering Integrated Circuit |

US20080200199 * | May 6, 2004 | Aug 21, 2008 | Philips Intellectual Property & Standards Gmbh | Method and Arrangement For Setting the Transmission of a Mobile Communication Device |

Classifications

U.S. Classification | 708/319 |

International Classification | G06F17/10, H03H17/02 |

Cooperative Classification | H03H17/0225 |

European Classification | H03H17/02E1 |

Legal Events

Date | Code | Event | Description |
---|---|---|---|

Apr 27, 2005 | AS | Assignment | Owner name: KONINKLIJKE PHILIPS ELECTRONICS, N.V., NETHERLANDS Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:RUNZE, GERHARD;REEL/FRAME:017020/0019 Effective date: 20031023 |

Aug 17, 2007 | AS | Assignment | Owner name: NXP B.V., NETHERLANDS Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:KONINKLIJKE PHILIPS ELECTRONICS N.V.;REEL/FRAME:019719/0843 Effective date: 20070704 Owner name: NXP B.V.,NETHERLANDS Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:KONINKLIJKE PHILIPS ELECTRONICS N.V.;REEL/FRAME:019719/0843 Effective date: 20070704 |

Dec 26, 2012 | FPAY | Fee payment | Year of fee payment: 4 |

Jan 28, 2016 | AS | Assignment | Owner name: ST WIRELESS SA, SWITZERLAND Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:NXP B.V.;REEL/FRAME:037624/0831 Effective date: 20080728 |

Feb 2, 2016 | AS | Assignment | Owner name: ST-ERICSSON SA, SWITZERLAND Free format text: CHANGE OF NAME;ASSIGNOR:ST WIRELESS SA;REEL/FRAME:037683/0128 Effective date: 20080714 Owner name: ST-ERICSSON SA, EN LIQUIDATION, SWITZERLAND Free format text: STATUS CHANGE-ENTITY IN LIQUIDATION;ASSIGNOR:ST-ERICSSON SA;REEL/FRAME:037739/0493 Effective date: 20150223 |

Rotate